bias in mapped forward/reverse read ratios

dvh · 09-22-2008, 05:09 AM

Hi,

We see a peculiar bias in the mapping of reads in forward and reverse directions to sites in human genomic DNA. Occurs with both RNAseq and DNAseq, and with both novoalign and maq mapping. 45bp SE GAII data.

We thought there should be a 50%/50% split of reads mapping to the genome in forward and reverse directions. The ratios of ./, or A/a etc (maq pileup output format) have huge variance, much more than expected by chance.

Has anyone else seen this? Is it something to do with solexa chemistry - eg adapter ligation, PCR steps?

It is relevant for SNP calling algorithms.

thanks

david

bioinfosm · 09-22-2008, 02:58 PM

Very interesting. We saw that once, but never pursued deeper. Can you be specific about the variance you see in fwd vs rev?

I will look up and see what we generally get..

sigusn · 10-01-2008, 03:43 AM

I've seen this too after using sequence capture for sample prepp. and it would be interesting to get an explanation for this. I got this reply from Illuminas Tech support :
"My understanding is that by using the Nimblegen sequence capture you enrich for particular regions but it also introduces a bias in the strandednes. We normally never see a bias with regards to strand.".
But I thought the DNA eluted from the seq.cap array would all be double stranded after the PCR amplification step, and therefor no bias should bee seen.

dvh · 10-01-2008, 04:06 AM

We see bias with both sequence capture DNA, and non sequence capture RNAseq. So its not the sequence capture.
david

Chipper · 10-01-2008, 04:53 AM

That sounds strange. Are there any particular regions where the bias occur and is it always the same strand that gets the higher read numbers?

clivey · 10-01-2008, 08:06 AM

Not sure I understand what you mean by bias ?

Is it global bias - so that say, 70% are F and 30% are R% ?

do you have a forwards and reverse strand coverage graph ?

you should see equal F and R globally, but if you make the F and R coverage plots you might see identical coverage peaks but offset by the average insert size of your template, because any fragment has a 50% chance of going onto the surface in either orientation during cluster prep.

grahamheap · 10-01-2008, 02:12 PM

Just thought i would give you an idea of what we are seeing in dvh's group. Attached to this post is a graph with some data from the phix174 control lane from two separate runs but using the same library. The data we see from other libraries (all kinds of sample preps - DNA, RNA, methylation) tends to be similar to this although the change is more pronounced than in phix, probably due to lower coverage.

cgb · 10-01-2008, 10:36 PM

that doesn't look right.

bioinfosm · 10-02-2008, 08:32 AM

Yes, definitely something strange happening there.. we see 55-45 fwd vs reverse mapping, but its kinda uniform across the reference..

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Separate forward and reverse coverage in Artemis	nivea	Bioinformatics	4	12-12-2011 04:20 AM
how to define a forward or reverse read file	poorphd	Illumina/Solexa	3	11-22-2011 01:34 PM
How to separate coverage of forward and reverse reads on same axis?	Kennels	Bioinformatics	6	05-04-2011 01:12 AM
Forward strand bias?	dawe	Bioinformatics	6	04-05-2010 08:00 AM
forward and reverse sequance	asankaf	General	5	05-27-2009 08:48 AM

09-22-2008, 05:09 AM	#1
dvh Member Location: london, uk Join Date: Jul 2008 Posts: 35	bias in mapped forward/reverse read ratios Hi, We see a peculiar bias in the mapping of reads in forward and reverse directions to sites in human genomic DNA. Occurs with both RNAseq and DNAseq, and with both novoalign and maq mapping. 45bp SE GAII data. We thought there should be a 50%/50% split of reads mapping to the genome in forward and reverse directions. The ratios of ./, or A/a etc (maq pileup output format) have huge variance, much more than expected by chance. Has anyone else seen this? Is it something to do with solexa chemistry - eg adapter ligation, PCR steps? It is relevant for SNP calling algorithms. thanks david

09-22-2008, 02:58 PM	#2
bioinfosm Senior Member Location: USA Join Date: Jan 2008 Posts: 482	Very interesting. We saw that once, but never pursued deeper. Can you be specific about the variance you see in fwd vs rev? I will look up and see what we generally get..

10-01-2008, 03:43 AM	#3
sigusn Member Location: Cambridge, MA Join Date: Aug 2008 Posts: 19	I've seen this too after using sequence capture for sample prepp. and it would be interesting to get an explanation for this. I got this reply from Illuminas Tech support : "My understanding is that by using the Nimblegen sequence capture you enrich for particular regions but it also introduces a bias in the strandednes. We normally never see a bias with regards to strand.". But I thought the DNA eluted from the seq.cap array would all be double stranded after the PCR amplification step, and therefor no bias should bee seen.

10-01-2008, 04:06 AM	#4
dvh Member Location: london, uk Join Date: Jul 2008 Posts: 35	We see bias with both sequence capture DNA, and non sequence capture RNAseq. So its not the sequence capture. david

10-01-2008, 04:53 AM	#5
Chipper Senior Member Location: Sweden Join Date: Mar 2008 Posts: 324	That sounds strange. Are there any particular regions where the bias occur and is it always the same strand that gets the higher read numbers?

10-01-2008, 08:06 AM	#6
clivey Member Location: Oxford Join Date: Jul 2008 Posts: 24	Not sure I understand what you mean by bias ? Is it global bias - so that say, 70% are F and 30% are R% ? do you have a forwards and reverse strand coverage graph ? you should see equal F and R globally, but if you make the F and R coverage plots you might see identical coverage peaks but offset by the average insert size of your template, because any fragment has a 50% chance of going onto the surface in either orientation during cluster prep.

10-01-2008, 10:36 PM	#8
cgb Member Location: Cambridge Join Date: May 2008 Posts: 50	that doesn't look right.

10-02-2008, 08:32 AM	#9
bioinfosm Senior Member Location: USA Join Date: Jan 2008 Posts: 482	Yes, definitely something strange happening there.. we see 55-45 fwd vs reverse mapping, but its kinda uniform across the reference..